openai / gpt-5.5

## Overview GPT-5.5 is OpenAI’s frontier text model for complex professional tasks. It reads long text context and optional images, then generates a text response that can include code, step plans, and structured JSON. It supports tool-based workflows like web search and file-based analysis when the host product enables them. It’s useful when you need one model to keep track of large projects and still produce clean, ready-to-use outputs. ## What you can build - Screenshot and document Q&A, like explaining UI errors from a PNG - Code review, refactors, and bug triage across large repositories - Data extraction from images, like tables, charts, and forms - Long-form drafting, like specs, reports, and internal docs - Structured text outputs, like JSON summaries for downstream systems - Research assistants that browse the web, then write a sourced brief ## Inputs - A required written request describing the task, question, or format you want. - Optional image uploads for visual understanding. Use PNG, JPG/JPEG, WebP, or non-animated GIF. Wiro supports up to 10 images per request. - Optional system-level instructions that set the assistant’s role, tone, and rules. Keep them short and specific. - Optional identifiers for chat history. Provide a user identifier and a session identifier when you want the model to continue a thread. - An optional reasoning effort choice. Higher effort tends to spend more tokens thinking on hard tasks. - An optional verbosity choice that controls how short or detailed the response is. - An optional toggle to let the model use web search during the answer. ## Outputs GPT-5.5 returns text output. That text can be plain prose, code blocks, or a structured JSON object if you ask for a specific schema. If you include images, the output typically includes descriptions, extracted details, and answers grounded in what’s visible. ## Limitations - GPT-5.5 outputs text only. It can accept images, but it does not output images, audio, or video. - The maximum context window is about 1,050,000 tokens, with up to 128,000 output tokens. Very large inputs may still need chunking. - The model has a fixed knowledge cutoff (December 1, 2025). Enable web search for newer facts. - Low-quality image inputs reduce accuracy. Blurry screenshots, tiny text, glare, heavy compression, or partial crops can break extraction. - The model may refuse requests for disallowed content, including instructions that enable wrongdoing. - Tool features depend on the host product. If the platform doesn’t expose a tool, the model can’t use it. ## Safety & compliance GPT-5.5 ships with OpenAI policy safeguards and safety monitoring. - It is evaluated against disallowed content categories such as hate, extremism, and self-harm. - OpenAI treats this launch as High capability in the Biological and Chemical domain, with extra safeguards active. - Cybersecurity requests are dual-use. The system may refuse high-severity offensive guidance. OpenAI also describes an identity-gated pathway for verified defenders in advanced cyber workflows. - If you connect GPT-5.5 to action tools (computer use, shells, or external systems), require explicit confirmation for high-risk actions like financial transactions or high-stakes communications.